huafu securities released a research report saying that since the ChatGPT released by OpenAi in 2022 pushed large-scale language generation models and AI to new heights, major technology giants around the world have embraced AIGC. Large models usually contain hundreds of millions of parameters and consume a lot of computing power. Server as the engine of computing power, the growth of computing power demand will usher in the rapid growth of AI server and server demand.

it is suggested to pay attention to: opportunities related to core links such as zhongke shuguang (603019.SH), hongxin electronics (300657.SZ), langke technology (300042.SZ) and component manufacturers in the server industry chain, such as chip manufacturer haiguang information (688041.SH), storage manufacturer langke technology (688008.SH), interface manufacturer yutai 300 (6882.S15.SH), server, server.

Huafu Securities views are as follows:

server as a computing engine, domestic and foreign demand resonance.

The opinions of the National Development and Reform Commission and other departments on the in-depth implementation of the "East Number and West Calculation" project to accelerate the construction of a national integrated computing network, which mentions that computing power is a new type of productivity in the digital economy era, with computing power empowering economic development as the main line, giving full play to the leading role of the national hub node of the national integrated computing network, and collaboratively promoting the "East Number and West Calculation" project, and build a national integrated computing network for networked scheduling. Since the ChatGPT released by OpenAi in 2022 pushed large-scale language generation models and AI to new heights, the world's major technology giants have embraced AIGC. Large models usually contain hundreds of millions of parameters and consume a lot of computing power. Server as the engine of computing power, the growth of computing power demand will usher in the rapid growth of AI server and server demand.

AI server industry chain includes chip CPU, GPU, memory DRAM and memory interface and HBM, local storage SSD,NIC, PCIe slot, heat dissipation and so on.

Server CPU architectures include X86, ARM, MIPS, and RISC-V. AI chip is the core of AI server computing power, which is specially used to process a large number of computing tasks in artificial intelligence applications. AI chip can be divided into GPU, FPGA, ASIC and NPU according to architecture. HBM, as a kind of memory product, has become standard for high-end GPU, which can be understood as the memory level corresponding to CPU or SoC. DDR and GPU chips originally on PCB board are integrated into SiP package at the same time, making the memory closer to GPU. Using HBM can greatly improve the communication bandwidth between DRAM and processors (CPU,GPU and other ASIC), thus alleviating the memory wall problem of these processors. The choice of server local storage includes HDD and SSD. The main hardware components of SSD include NANDFlash, main control chip and DRAM. The core software is firmware of enterprise SSD. Data center SSD is no longer a hard disk, but a small system with processing, caching, calculation and security protection. SSD penetration rate is expected to gradually increase.

judging from the composition of servers and AI servers, chips and storage are the cores, which determine the computing power and broadband size of AI servers.

in the traditional general-purpose server, the 2 xIntelSaphireRapidsServer, which costs US $10424, accounts for about 17.7 percent of CPU cost and more than 50% of memory and hard disk. In AI servers, CPU accounts for only 1.9 percent and GPU accounts for 72.6 percent of the $268495 NvidiaDGXH100. The value of memory increased, but the share fell to about 4.2 percent. AI servers are significantly more valuable than general-purpose servers.

Risk Warning: Geopolitical risks, AI development less than expected, customer certification and product validation less than expected risks.